Large-scale Experimentation with Network Abstraction for Network Configuration Management
1. Large scale RINA Experimentation on FIRE +
Large-scale Experimentation with Network Abstraction
for Network Configuration Management
Dr. Sven van der Meer, NM-Lab, Ericsson
2. ntroduction
opic:
Management of multi-layer converged service provider networks
bjectives, demonstrate that
RINA leads to simplification of management definitions, implementations
Coordinated management can evolve
– Complex workflows to a strategy-oriented management task
Single manager can simultaneously perform
– Adequate number of strategies
Manager can be scaled out/up and scaled in/down
– In case a single manager is insufficient
Large-scale RINA Experimentation on FIRE+
3. xperiment Design
1 experiment
– Create network, create and validate nodes, validate network, generate
configuration script, generate network report
Run for 24 networks
– From tiny (rumba-2-nodes) to extra large (metro2110)
Run on 6 machines on 3 hardware platforms
– Raspberry PI, Windows/Cygwin, different UNIX machines including large server
Using 1 management strategy
– Consisting of 2 core policies: network and node management
– Some secondary policies: context policy parameters, and templates
– Using identical software installation and network scenarios
Large-scale RINA Experimentation on FIRE+
5. Main Statement
his experiment provides substantial experimental evidence that
Consequent application of RINA abstractions
Leads to significant improvements
In configuration management
Especially for
– performance,
– management software complexity
– automation
Large-scale RINA Experimentation on FIRE+
6. Explored Abstractions
Large-scale RINA Experimentation on FIRE+
NA
Separate mechanism
from policy
Mechanism/Policy are
relative
1 application protocol
(CDAP)
Management is
monitoring and repair
(PtP) DIF and DAF
ARCFIRE Initial
• Management
requires initial
configuration
• Management is
relative (term and
activities)
• Network: set of
nodes, DIFs, IPCPs in
DIFs
ARCFIRE Execution
• Nodes can be isolate
(for create / validate
• Network configuratio
requires only node
specifications
• 4 graphs seem to be
important to show a
network
8. DMS – Components
APEX – (adaptive) policy engine
– https://ericsson.github.io/apex-docs/ and in ONAP
– With D-MIM – distributed object (shared memory) management
Apex Policy Builder (APB) – develop and test strategy, create artifacts
– Plus set of clients
DMS – deploy and run strategy
Management Strategy
– Set of policies specified in form of a policy model
– Core: policy to handle networks, policy to handle nodes
– Secondary: policies handling context information, policy parameters, templates
– Binaries: scripts to create artifacts, measurement client
Large-scale RINA Experimentation on FIRE+
14. KPI AND MAIN RESULTS
Large-scale RINA Experimentation on FIRE+
15. DMS – KPIs
Speed – of the management strategy
Scale – requirements to scale out the DMS
Time and cost of scaling
Touches – required human interference
Strategy complexity
– Runtime: number of operations
– Code: counted lines of code
Degree of automation
– 0: not automated at all
– 100: fully automated
Large-scale RINA Experimentation on FIRE+
23. OTHER RESULTS – VISUALISATION
Large-scale RINA Experimentation on FIRE+
24. Other Experiment 1 Results
Network visualization
– 4 graphs, generated by the management strategy
• Network (nodes) graph: isolated nodes with their DIF structure
• Point-to-Point (PtP) graph: node connectivity (commonly named “topology”)
• DIF graph: all DIFs in a network, and their connections
• Network (IPCP) graph: all IPCPs in their DIFs per node
– Node finger print using onion diagrams
– Network 2-D view with onion diagrams
– Network 3-D view using onion diagrams
RINA DAF Model (Application Model)
– Detailed model of a distributed application, as executables
– With infrastructure
– With RINA distributed infrastructure
– Facilitiy and DMS model
Large-scale RINA Experimentation on FIRE+